Implementing a Hashmap in C++ :: hashing function for templated data type

Question

I've been using STL's unordered_map recently and while it seems to work nicely I don't quite understand how the hashing function works given that the data type is given as a template parameter. In an effort to understand this data structure more thoroughly, I implemented my own little Hashmap class in C++ :

Hashmap interface:

#ifndef _HASHMAP_H_
#define _HASHMAP_H_

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <iostream>
#include <vector.h>


//Beginning of Hashmap class definition

template <class Key, class Value>
class Hashmap{
private:

int mappedElementCount;



public:
explicit Hashmap();
virtual ~Hashmap();


virtual void test();

virtual int hash(Key*);

int* getSize();

void putKVPair(Key*,Value*);

void clearMap();


//When we use these methods, we'll want a linear vector of keys and values to 
    //iterate over, so vector is good here
std::vector<Key>* getKeys();
std::vector<Value>* getValues();

}; //end hashmap class definition
#endif /*_HASHMAP_H_*/

Hashmap implementation:

#include "Hashmap.h"

template<class Key,class Value> Hashmap<Key,Value>::Hashmap(){
mappedElementCount = 0;
}
template<class Key,class Value> Hashmap<Key,Value>::~Hashmap(){
printf("\nDestroying the base Hashmap object!\n");
}

template<class Key,class Value> void Hashmap<Key,Value>::test(){
printf("The size of our Key is %i and the size of our Value is
    %i\n",sizeof(Key),sizeof(Value));
}


template<class Key,class Value> int Hashmap<Key,Value>::hash(Key* k_ptr){

    unsigned int hashval;

    /* we start our hash out at 0 */
    hashval = 0;

        //TODO: How do we generate a hash signature when we don't know what data type 
        //we're going to be working with?

    return hashval % mappedElementCount;

}

template<class Key,class Value> std::vector<Key>* Hashmap<Key,Value>::getKeys(){
//TODO: prepare a vector initialized with all Key objects and return it here
return keys;    
}

template<class Key,class Value> std::vector<Value>* Hashmap<Key,Value>::getValues(){
//TODO: prepare a vector initialized with all Value objects and return it here
return values;  
}

template<class Key,class Value> int* Hashmap<Key,Value>::getSize(){
return &mappedElementCount;
}

template<class Key,class Value> void Hashmap<Key,Value>::putKVPair(Key* k, Value* v){
    //TODO: implement hashing of the key object k to determine
    //the address of the value object v

    //first step, generate a hash from our key
    int tempHash = hash(k);

       //TODO: store the Value at an address given by or influenced by tempHash

    //If all was successfully completed, increment the mapped records counter
    mappedElementCount++;
}



template<class Key,class Value> void Hashmap<Key,Value>::clearMap(){
//TODO: implement a cascading chain of deallocation of stored objects within the 
    //hashmap
//MAYBE-- only if we create new objects rather than just mapping reference 
    //associations,
//which is really the goal here...  In the latter case, just empty the Hashmap 
    //itself
}

One possible OOP method of addressing this problem is to use Hashmap as a base class and provide derived classes that have known Key data types, such as the following Stringmap:

Stringmap interface:

#ifndef _STRINGMAP_H_
#define _STRINGMAP_H_

#include "Hashmap.h"

template <class Value>
class Stringmap:public Hashmap<std::string,Value>{
private:

public:
//Con/de 'structors
explicit Stringmap();
~Stringmap();

//Here we know our Key will be of type std::string
//so we can generate our hash sig by char values
    //Override hash from the base class
int hash(std::string*);

//override test from base class
void test();


};
#endif /*_STRINGMAP_H_ def*/

Stringmap implementation:

#include "Stringmap.h"

template<class Value> Stringmap<Value>::Stringmap():Hashmap<std::string,Value>(){

}
template<class Value> Stringmap<Value>::~Stringmap(){
printf("\nDestroying the derived stringmap object!\n");
}

template<class Value> void Stringmap<Value>::test(){
printf("The size of our Value is %i\n",sizeof values[0]);
}

template<class Value> int Stringmap<Value>::hash(std::string* str_ptr){

    unsigned int hashval;

    /* we start our hash out at 0 */
    hashval = 0;


    /* for each character, we multiply the old hash by 31 and add the current
     * character.  Remember that shifting a number left is equivalent to
     * multiplying it by 2 raised to the number of places shifted.  So we
     * are in effect multiplying hashval by 32 and then subtracting hashval.
     * Why do we do this?  Because shifting and subtraction are much more
     * efficient operations than multiplication.
     */
    for(int i=0;i<str_ptr->length();i++) {
        hashval = (*(str_ptr))[i] + ((hashval << 5) - hashval);
    }

    /* we then return the hash value mod the hashmap size so that it will
     * fit into the necessary range
     */
    return hashval % (*(Hashmap<std::string,Value>::getSize()));

}

So the question is as follows: is it possible to create a hash signature when the data type to be hashed is currently unknown? If so, how? Looking at the std::hash docs, it appears that the C++ standard simply defines a hash function for each primitive data type and also for T* (for any type T)... What's missing is how this hashing is implemented for the given primitive data types and, more to the point, how it is implemented for the generic T*. I suppose I could just call hash(Key) and hope for the best, but it would be nice to understand what's going on behind the scenes.

thanks, CCJ

Answer 1

std::unorderd_map takes 2 explicit template parameters (Key, and Value) and also has a bunch of hidden template parameters, of which the Hash function is defaulted to std::hash<Key>.

This STL hash function std::hash<Key> takes a Key and returns a std::size_t. It is already specialized for all the integral types and std::string. From this reference site

The hash template defines a function object that implements a hash function. Instances of this function object define an operator() that:

1. Accepts a single parameter of type Key.
2. Returns a value of type size_t that represents the hash value of the parameter.
3. Does not throw exceptions when called.
4. For two parameters k1 and k2 that are equal, std::hash()(k1) == std::hash()(k2).
5. For two different parameters k1 and k2 that are not equal, the probability that std::hash()(k1) == std::hash()(k2) should be very small, approaching 1.0/std::numeric_limits::max().

The hash template is both CopyConstructible and Destructible. The unordered associative containers std::unordered_set, std::unordered_multiset, std::unordered_map, std::unordered_multimap use specializations of the template std::hash as the default hash function.

The reference ends with this quote:

** The actual hash functions are implementation-dependent and are not required to fulfill any other quality criteria except those specified above. **

So you can look at the implementation of your system, but that won't guarantee anything for other systems' implementation.

Answer 2

There's a std::hash<T> template which is specialized for a variety of types, and which you can specialize for your own types.

By default, std::unordered_map<T> simply delegates hashing to std::hash<T> (or you can specify a different hash function as a template argument).

Thus std::unordered_map does not need to know anything at all about the mechanics of hashing.

As to how std::hash is implemented, that's not specified. However, I think it's reasonable to assume that any decent compiler would provide a good-quality implementation. One gotcha to bear in mind is that std::hash<char*> doesn't hash the C string, it only hashes the value of the pointer (been there :))